Hardware Advice: Jetson Orin NX (16GB) vs. AGX Orin (64GB) for Triple 100 FPS Video Processing

Robotics & Edge Computing Jetson Systems Jetson AGX Orin
1

Hi everyone,

I’m looking for some guidance regarding the hardware requirements for a computer vision project. I need to process video streams from three cameras simultaneously to detect surface defects on a cable moving at 160 m/min.

The Setup:

  • Cameras: 3 units positioned at 120° intervals around the cable.

  • Frame Rate: 100 FPS per camera (providing a safety overlap, whereas 67 FPS would be the strict minimum for continuous coverage).

  • Total Throughput: 300 FPS total to be processed in real-time.

The Dilemma: I am hesitating between the NVIDIA Jetson Orin NX (16GB) and the Jetson AGX Orin (64GB).

Based on my initial assessment, the Orin NX (100 TOPS) seems like it might be the baseline for this kind of work. However, general feedback and online forums often point toward the AGX Orin (275 TOPS) for multi-stream high-speed processing.

My Questions:

  1. Is the Orin NX 16GB realistically capable of handling the inference and pre-processing of three 100 FPS streams without significant latency or dropped frames?

  2. Given the cable speed (160 m/min), the processing must be near-instantaneous to be useful. Would the Orin NX’s memory bandwidth be a bottleneck compared to the AGX Orin?

  3. If the Orin NX is too tight, is the AGX Orin the only logical step up, or is there a “middle ground” solution I should consider?

I would love to hear from anyone who has deployed similar multi-camera high-FPS setups on the Orin platform.

Thanks in advance for your help!

2

Hello @Florian-Gonzalez!

Based on the title and content of your topic, it looks like it may receive better visibility and feedback in a different category. We took the liberty of moving it for you.

If this was an incorrect assessment, please send me a direct message.

Disclaimer: this moderation suggestion and message were generated with AI assistance.

4

Hi,
Please share more information about the camera source running 100fps, like which interface is used(PCIe, USB, CSI, …), and what the resolution is. Do you use v4l2 interface to capture frame data?

5

Hi,
I will be using three Basler ace 2 cameras with the 5GigE color interface. Regarding the resolution, I aim for the highest native resolution supported by the sensor at 100 FPS to ensure maximum defect detection accuracy for the cable surface. The cameras will use the 5GigE interface, and I am planning to use a specialized NIC or a robust switch to handle the combined throughput of all three devices.
Concerning the capture method, I have not yet decided on V4L2 because I am considering using the Basler pylon SDK for better hardware integration, though I am open to suggestions like DeepStream or GStreamer for optimal performance on the Jetson platform.
Given the high bandwidth of three 5GigE streams, do you believe the Orin NX’s PCIe and networking overhead would be a bottleneck for this specific real-time application?

6

Hi,
If you use Orin NX, you may need to design the custom board. The developer kit(Orin NX module + Orin Nano carrier board) has PHY with 1Gbps. The bandwidth looks insufficient for the camera.

Also suggest consult with the vendor whether they have validated three Basler ace 2 cameras in 100fps on either Jetson platform. Generally we suggest 60fps for camera inputs.

7

Hi,

Thank you for your feedback. Regarding the bandwidth concerns, I have decided to move forward with the Jetson AGX Orin 64GB Developer Kit instead of the Orin NX.

To handle the high data throughput, I plan to use a dedicated PCIe card with multiple ports (supporting up to 10Gbps per port). This will allow the data streams to flow directly from the cameras to the NVIDIA module via the PCIe bus, bypassing the 1Gbps bottleneck of the standard carrier board.

As for the frame rate, I am targeting 100 FPS to ensure a minimum of 20% overlap between consecutive images. Given that the cable moves at 160 m/min and I need to detect defects as small as 1 mm, this overlap is important to ensure that no part of the surface is missed and to provide better consistency for the AI model.

Given these requirements (3 cameras, 300 FPS total, 1mm defect sensitivity), do you agree that this AGX Orin + PCIe NIC setup is the most suitable architecture, or would you suggest any further optimizations for such high-speed inspection?

8

Hi,

Based on vendor recommendations, I am now planning to use three ELM-12-4.0-18-C-E modules (integrated lens + ECC-1C controller) with my Jetson AGX Orin 64GB.

These modules are powered and controlled via USB. I will need to connect three of them simultaneously to the Jetson’s USB ports.

Questions :

Can the USB 3.2 ports on the AGX Orin DevKit provide a stable 5V supply for three modules at once (approx. 500mA each)?

Are there any known limitations when using multiple USB-to-UART controllers simultaneously on JetPack?

9

Hi,
Please check section 2.1 in Board Specification:

Jetson Download Center | NVIDIA Developer

It is 5V/2A for the type-A ports. However, performance may drop if connecting multiple devices to the same USB roothub. Would suggest shift one device to the type-C port.